version
stringAlways "1" for v1.
Delx Continuity Capsule v1
One JSON object between death and cold start.
An open handoff format. Free to emit, free to read, versioned, and published as a JSON Schema so no agent has to guess the shape.
What is this
Open JSON Schema for agent handoffs.
Delx Continuity Capsule v1 is an open format for what an agent leaves behind: goal, done, next, blockers, do_not, and refuted hypotheses. Any agent from any vendor can emit or consume it. This is not a local coding skill like "Context Capsule" — it is a published JSON Schema, not a public feed.
JSON Schema: https://api.delx.ai/schemas/continuity-capsule-v1.json
goal
stringWhat this run was trying to achieve.
done
stringWhat is finished and verified — not what was attempted.
next
stringThe single next action, concrete enough to start cold.
blockers
stringWhat is stuck, and on whom or what.
do_not
stringConstraints the next session must not violate.
refuted★
stringHypotheses already ruled out, with why. The field that saves the most tokens.
receipts
string[]Pointers to evidence: commits, deploy ids, run urls.
written_by
stringWhich agent sealed this trail.
written_at
stringISO timestamp; the resume surfaces the age.
ttl_days
string | intHow long the sender considers this useful.
handoff_to
stringOptional: the intended receiver.
Example
A real capsule.
This is what a mid-migration handoff looks like when it is written to be read.
capsule.json
{
"version": "1",
"goal": "Migrate the billing module to the new client",
"done": "Adapters written, unit tests green, staging deploy verified",
"next": "Swap the callsites in checkout, then delete the shim",
"blockers": "Staging secret STRIPE_KEY_V2 is missing",
"do_not": "Do not delete the shim before the callsites move",
"refuted": "Batching the migration fails — the old client is not idempotent",
"receipts": ["repo@a1b2c3d", "staging deploy 2026-08-08T11:04Z"],
"written_by": "acme-refactor-bot",
"ttl_days": 14
}Validator
Validate yours.
Nothing leaves your browser. The rules below mirror the published schema and the write guards the API enforces.
Paste a capsule above, or load the example.
Rules mirrored from the published schema; the API stays the authority.
JSON Schema
The schema.
Fetch it, generate types from it, or validate against it in CI.
curl
curl https://api.delx.ai/schemas/continuity-capsule-v1.jsonA capsule is stored as hive.* memory keys on the sending session and returned to the next session under the same stable agent_id. It is not a public feed, not a shared board, and not indexed anywhere. Fleets are the one exception, and only for members who joined with an invite token.
FAQ
Direct answers.
What is the difference between Continuity Capsule and Context Capsule?
They are different products. "Context Capsule" refers to local coding skills used in some development environments. Delx Continuity Capsule v1 is an open, published JSON Schema for agent handoffs — it is not a public feed or a local skill.
Is the Continuity Capsule format open?
Yes. The JSON Schema is published at a stable URL, versioned, and free to implement. Any agent from any vendor can emit or consume a capsule without using Delx to store it.
What is the refuted field for?
It records hypotheses that have already been ruled out and why. It is the highest-value field in a handoff because it prevents the receiving session from spending its context re-deriving a dead end the sender already paid for.
What happens if a capsule contains a secret?
The write is refused before storage. Credential-shaped values are detected and rejected with an explanation that never echoes the value itself, and nothing is logged.
How large can a capsule be?
Each string field is capped at 4,000 characters and a session can hold up to 16 hive keys. The limits exist so a trail stays readable by the next agent rather than becoming a transcript dump.